|Publication number||US6697316 B2|
|Application number||US 09/847,775|
|Publication date||Feb 24, 2004|
|Filing date||May 1, 2001|
|Priority date||May 1, 2001|
|Also published as||US20020172131|
|Publication number||09847775, 847775, US 6697316 B2, US 6697316B2, US-B2-6697316, US6697316 B2, US6697316B2|
|Inventors||Geoffrey W. Burr|
|Original Assignee||International Business Machines Corporation|
|Export Citation||BiBTeX, EndNote, RefMan|
|Patent Citations (10), Non-Patent Citations (16), Referenced by (39), Classifications (8), Legal Events (7)|
|External Links: USPTO, USPTO Assignment, Espacenet|
The invention relates to the field of holographic data storage, and more generally, to the problem of correcting errors that arise from the misregistration of detectors (e.g., pixels) and coherent beams of optical radiation.
The ever increasing demand for readily accessible information has been enabled by the increasing data storage capacity of computers and the decreasing cost of storing this data. Magnetic storage media have played a prominent role in the Information Age, and more recently, optical data storage technologies have become commonplace. However, both magnetic and conventional optical data storage technologies, in which individual bits are stored as distinct markers on the surface of the recording medium, are approaching physical limits beyond which the storage of even more information is impractical. Storing information throughout the volume of a medium, rather than just on its surface, appears to be a high capacity alternative to currently used information storage technologies. Holographic data storage is one such volumetric approach to data storage.
As shown in FIG. 1, with holographic data storage, an entire “page” of information in the form of bit images is recorded at one time as an optical interference pattern within a holographic storage medium 10. Lithium niobate is one photosensitive optical material commonly used in holographic storage applications. The input data to be stored in the medium 10 is optically prepared by passing an input beam 14 of that light (e.g., from a laser, not shown) through a spatial light modulator (SLM) 18. The SLM 18 functions as a “pixelated” input device and typically includes a liquid crystal panel similar to those used in laptop computers. Individual pixel elements 22 within the SLM 18 may be turned on or off (although shades of gray are also possible), as illustrated in FIG. 1 by the SLM's white and dark elements, respectively. In this manner, the input beam 14 is spatially modulated to form an object beam 26 containing a page of information determined by the particular on/off configuration of the pixel elements 22 of the SLM 18; the object beam 26 may be thought of as containing an array of individual beams (not shown in FIG. 1) corresponding to SLM pixel elements 22. This information-containing object beam 26 may then be directed through one or more optical elements 30 such as a lens and into the holographic storage medium 10. The information contained in the object beam 26 can be recorded in the storage medium 10 by intersecting the object beam with a second beam 34 of coherent light (e.g., from a laser, not shown) known as the reference beam. The reference beam 34, which is of the same wavelength as the object beam 26, is designed to be easy to reproduce, e.g., it may be simply a collimated beam having a planar wavefront. Intersecting the object beam 26 and the reference beam 34 in the storage medium 10 produces an optical interference pattern that results in chemical and/or physical changes in the medium itself, such as a change in absorption or refractive index, thereby producing a grating within the storage medium. This optical interference pattern, recorded as physical changes to the material within the storage medium 10, contains the information represented by the corresponding on/off (or gray level) configuration of the SLM pixel elements 22.
Information may be retrieved from the storage medium 10 as illustrated in FIG. 2. The reference beam 34 is directed into the storage medium 10 which has now been physically altered to contain a grating corresponding to the interference pattern generated by the intersection of the object beam 26 and the reference beam 34. When the storage medium 10 is illuminated in this manner, some of the reference beam 34 is diffracted by the grating such that a reconstructed object beam 38 emerges from the storage medium 10. This reconstructed object beam 38 contains the same information (and propagates in the same direction) as the object beam 26. The reconstructed object beam 38 may then be imaged through one or more optical elements 42 (e.g., lens) and onto a grid or array 44 of discrete detector elements 48 or pixels (e.g., an array of nominally square pixels in a charge coupled device or CCD), so that the page of information (or “data page”) originally modulated onto the object beam 26 by the SLM pixel elements 22 can be retrieved. Alternatively, information may be retrieved by illuminating the storage medium 10 with a reference beam (not shown) that is phase-conjugate to the reference beam 34 (see J. Ashley et al., “Holographic data storage”, IBM J. Res. Develop., vol. 44, no. 3, May 2000, pp. 341-368). This phase conjugate reference beam is diffracted by the grating within the storage medium 10 to construct a phase-conjugate object beam. A beamsplitter and a polarization element maybe used to direct this phase-conjugate object beam onto the detector array 44. The relationship between a beam and its phase-conjugate is that they have exactly the same spatial wavefront, but propagate in opposite directions (much like a movie played in reverse).
A large number of interference gratings can be formed within a single storage medium 10 by altering the angle between the object beam 26 and the reference beam 34, or alternatively, by using other wavelengths for the object and reference beams. In this manner, a large number of data pages may be stored in the storage medium, and any particular data page can be read out independently of other data pages. The theoretical limit to this volumetric approach to data storage and retrieval has been estimated to be on the order of tens of terabits per cubic centimeter. (See J. Ashley et al., supra.) Holographic data storage also offers the prospect of faster access times, since laser beams can be moved rapidly (unlike actuators used in disk drives), and the output wavelength of certain kinds of lasers can be rapidly and accurately tuned.
To retrieve data from the storage medium 10 without error, the data page represented by the complex optical wavefront of the reconstructed object beam 38 would ideally be imaged onto the pixels 48 of the detector array 44 such that each of the individual beams (not shown in FIGS. 1 and 2) corresponding to an individual SLM pixel element 22 fell onto a single pixel 48 in the detector array 44. That is, the reconstructed optical beam 38 would be imaged onto the array 44 such that there would be a one-to-one correspondence between the SLM pixel elements 22 and the detector pixels 48. In practice, such a high degree of correlation is difficult to achieve. Misfocusing of the detector array 44 (e.g., with respect to the focal plane of the optical element 42) as well as optical aberrations in the imaging system lead to a situation in which energy that was intended for a particular pixel 48 lands on one or more adjacent pixels. This misregistration of the pixels 48 with the individual beams within the reconstructed object beam 38 results in interpixel interference or crosstalk. Interpixel interference may lead to errors in the retrieved data when signals from the pixels 48 are read out to, for example, a processor or computer (as in FIG. 3 below). However, in current holographic systems, some stressing of the physical components of the holographic system beyond the point of ideal imaging may be acceptable if coding and signal processing techniques can be employed to reduce the bit error rate (BER) to a sufficiently low level.
As discussed above, each portion of the data-bearing optical wavefront intended for an SLM pixel element 22 should ideally land on a single detector pixel 48. Both the SLM 18 and the output detector array 44 typically contain pixels on a nominally square grid having a grid-spacing or pitch δ. If the pixel pitches of the input and output devices are different, the optical imaging system is ideally designed to contain the appropriate magnification. However, the active area of the detector pixels 48 may be smaller than the pixels 48 themselves, giving rise to the “fill factor” of the pixel, measured either along the side of the pixel or as a fraction of the area used: active area/δ2.
To maximize storage density, the holographic recording is done at a point where the data-bearing beam is tightly focussed—often an aperture is used to prevent other portions of the storage medium from being inadvertently exposed. The smaller this aperture, however, the more the resulting pixel images are blurred at the detector array 44. The optical system is characterized by a point spread function (PSF), which is the size of the (blurred) spot found at the output image plane when a point source is placed at the input plane (or alternatively, when a hologram of this same wavefront is reconstructed). Since the aperture is finite in size, the PSF will have periodic nulls or zeroes, and the spacing between these nulls will scale inversely with the limiting aperture at the medium. In principle, crosstalk between pixels could be entirely avoided by the combination of making each pixel very small in area followed by choosing an aperture such that the spacing between the peak and the first null of the PSF is identical to the pitch of the detector pixel 48. This aperture is termed the Nyquist aperture, and it is the smallest aperture for which sufficient information can be passed from the SLM 18 to the output array 44 to guarantee the reliable retrieval of information. This aperture may be defined by the dimensions of the storage medium 10 or by a component (not shown) positioned between the SLM 18 and the detector array 44. Unfortunately, the number of photons detected at the detector array 44 must be sufficient to overcome thermal noise at the detector array, and the benefits of a detector array arising from low-crosstalk are far outweighed by the loss of signal arising from the very small fill-factor. (See M-P. Bernal et al., “Balancing interpixel cross talk and detector noise to optimize areal density in holographic storage systems”, Applied Optics, vol. 37, no. 23, pp. 5377-5385, 1998).
Linear signal processing techniques such as equalization, Wiener filtering, and partial-response-maximum-likelihood have had great success in deblurring the temporal 1-D channels found in conventional storage and communications systems. (See J. W. M. Bergmans, Digital baseband transmission and recording, Kluwer Academic Publishers, Boston, 1996.) This success led some investigators to propose signal processing schemes for page-oriented data storage that extend existing 1-D algorithms to 2-D but retain a simple linear channel model. (J. F. Heanue et al., “Signal detection for page-access optical memories with intersymbol interference”, Applied Optics, vol. 35, no. 14, pp. 2431-2438, 1996; V. Vadde et al., “Channel modeling and estimation for intrapage equalization in pixel-matched volume holographic data storage”, Applied Optics, vol. 38, no. 20, Jul. 10, 1999; and B. M. King et al., “Parallel detection algorithm for page-oriented optical memories”, Applied Optics, vol. 37, no. 26, Sep. 10, 1998.) Their reasoning was that by correcting the blur introduced by diffraction, it should be possible to work with smaller spatial apertures and thus higher areal densities.
However, the performance improvements achieved in holographic data storage systems using these techniques have been less dramatic than those achieved in 1-D systems. One reason for this is that the detection of the intensity of a coherent optical wavefront is an inherently nonlinear process, in which the electric fields (not the intensities) of two or more beams constructively or destructively interfere with each other. The resulting optical intensity at any given point is given by summing the electric fields from all the beams arriving at that point and squaring this sum. The intensity at each point of a pixel may depend (at least in part) on electric fields from signals that would have landed onto nearby pixels in the absence of misregistration. Each detector pixel integrates intensity over that pixel's area to generate an output signal, in which the intensity at each point is given by summing the square of the total electromagnetic field at that point. Thus, in the absence of interference effects arising from coherent light, intensities from two or more different beams would simply add. With coherent light, however, deconvolving the intensity data from a pixel array to establish what the intensity data would have been in the absence of interference effects arising from misregistration is fundamentally a nonlinear process.
The problem of misregistration has typically been addressed through careful (but costly) engineering. The optical components between the SLM 18 and the detector array 44 can be configured to minimize optical distortion and magnification error, but simultaneously achieving high storage density and compact system size can be difficult. Alternatively, phase-conjugate readout can be used to reduce the impact of aberrations in most but not all of the optical system (F. Zhao et al., “High density phase-conjugate holographic memory with phase-only image compressors”, Optical Memory and Neural Networks, vol. 6, no. 4, pp. 261-264, 1997). In either case, however, precise optomechanical mounting and some degree of lateral position feedback are necessary to achieve and maintain alignment. In practice, small data pages can almost be perfectly pixel-matched, i.e., the problem of misregistration can be, for all practical purposes, eliminated. However, with large megapixel or megapel (1024×1024) data pages, a significant fraction of the signal-to-noise (SNR) budget can be consumed by even a small amount of optical distortion: In practice, one part of a data page must be slightly misaligned in order to bring another part of the data page into perfect alignment. Thus, there remains a need for a method of compensating for the errors resulting from the misregistration of a data page with an array of detector pixels.
One aspect of the invention is a processor-implemented method of determining the intensities of beams of coherent electromagnetic radiation, in which the beams form an array. The method comprises providing an array of detectors and imaging the array of beams onto the array of detectors to generate output signals from the detectors. The output signals are detected to record respective output signal levels, and the misregistration between the detectors and the beams is determined. A processor is used to correct the recorded output signal levels by compensating for electromagnetic interference effects at the detectors arising from the misregistration, so that the corrected output signals levels are proportional to those that would be recorded in the absence of misregistration. In a preferred implementation, the detector array is a two-dimensional array of pixels. The array may include at least (or exactly) 1024×1024 pixels, or it may include at least (or exactly) 1024×768 pixels. In one preferred implementation, the coherent radiation includes laser light.
Determining the misregistration may advantageously include determining a set of static misregistrations between the detectors and the beams, determining any dynamic misalignment between the array of detectors and the array of beams, and then determining a set of total misregistrations between the detectors and the beams from the dynamic misregistration and the set of static misregistrations. In a preferred implementation, determining the dynamic misalignment between the array of detectors and the array of beams includes identifying the location of fiducial images on the detector array. The imaging of the array of beams preferably includes directing a beam of laser light through a holographic storage medium to reconstruct the array of beams. In one preferred implementation, compensating for electromagnetic interference effects includes determining, for each detector, the convolution of the optical point spread function (PSF) of beams in the array and the detector's area. The compensating preferably includes performing iterative calculations in a first spatial dimension for detectors along a given row aligned along the first spatial dimension, and advantageously further includes performing iterative calculations in the first spatial dimension for detectors along additional rows aligned along the first spatial dimension. The compensating preferably further includes performing iterative calculations in a second spatial dimension for detectors along rows aligned with the second spatial dimension.
Another aspect of the invention is a processor-implemented method of correcting errors arising from interpixel cross talk in the detection of a reconstructed holographic data image, in which the data image includes a two dimensional grid of bit images arranged in rows and columns. The method includes providing a detection device having a grid of detector pixels, in which the detector pixels correspond to and detect the bit images. The data image is directed onto the detector pixels by passing coherent electromagnetic radiation through a holographic storage medium, and output signals are measured from the detector pixels. Any misregistration between the pixels and the bit images is determined, and a processor is used to correct the measured output signals to compensate for signal strength lost to or gained by nearby pixels through interference effects resulting from the misregistration, thereby producing a calculated data image that is substantially corrected for interpixel crosstalk. In a preferred implementation, the misregistration includes determining any dynamic misalignment between the image grid and the pixel grid, and determining any static misregistrations between the bit images and the pixels.
Correcting the measured output signals may advantageously include (a) for each pixel in a given row of pixels, adding to the output signal measured at that pixel any signal lost through interference effects to neighboring pixels in the given row, (b) for each pixel in the given row, subtracting from the output signal measured at that pixel any signal gained through interference effects from neighboring pixels in the given row, and (c) applying (a) and (b) to additional rows of pixels, thereby producing a data image of revised output signals. Correcting the measured output signals may further include (d) for each pixel in a given column of pixels, adding to the revised output signal for that pixel any signal lost through interference effects to neighboring pixels in the given column, (e) for each pixel in the given column, subtracting from the revised output signal for that pixel any signal gained through interference effects from neighboring pixels in the given column, and (f) applying (d) and (e) to additional columns of pixels. Analogous methods may be employed, in which columns are processed before rows.
A preferred implemenation includes iteratively processing the output signals from the detector pixels along a first dimension to produce an intermediate page of intensities, and processing the intermediate page intensities along a second dimension to produce the calculated data image. In one preferred implemenation, patterns of discrete bit images are imaged onto the detector pixels to determine the extent of the interference effects resulting from misregistrations between detector pixels and their respective bit images. In a preferred implementation, output signals are iteratively processed back and forth along each row to generate respective revised data signals, and output signals are iteratively processed back and forth along each column to generate respective revised data signals, in which only those revised data signals for which pixel offsets are less than or equal to one-half of the pixel spacing are retained. In another implementation, output signals are iteratively processed back and forth along each row to generate respective revised data signals, and output signals are iteratively processed back and forth along each column to generate respective revised data signals, in which, for at least some detector pixels, an average of the revised data signals is calculated. The method advantageously comprises including blank rows and columns in the data image to facilitate correcting of the measured output.
FIG. 1 illustrates how a spatial light modulator (SLM) can be used in combination with two coherent beams of light to store a data page of information in a photosensitive material.
FIG. 2 illustrates how the information in the data page recorded in FIG. 1 can be retrieved using a coherent beam of light.
FIG. 3 shows optical beams in a pixelated data page (corresponding to pixels in the SLM) propagating towards pixels of a detector array, in which the optical beams are shifted to the right of the detector pixels.
FIG. 4 shows how electric fields from the optical beams of FIG. 3 may interfere with each other at the detector pixels.
FIG. 5 shows electric fields from the optical beams of FIG. 3 interfering at the detector pixels in the case that one of the optical beams has been turned off.
FIG. 6 shows electric fields from optical beams interfering at detector pixels in the case that one of the optical beams has been turned off, and with the optical beams shifted to the left of the detector pixels.
FIG. 7, which includes FIGS. 7A, B, C, and D, illustrates how the preferred order in which signals from the detector pixels are processed can vary depending upon how the pixelated data page is aligned with the detectors in the detector array.
FIG. 8, which includes FIGS. 8A, B, and C, shows a 9×9 pixel pattern being imaged from a pixelated data page to a detector array (A) under perfect conditions, (B) with a half pixel offset in both the x and y dimensions, and (C) after processing the output signals from the detectors with a shift-compensation algorithm.
FIG. 9 shows how fiducial marks may be incorporated into a pixelated data page to determine misregistration between the data page and the pixels of a detector array.
FIG. 10, which includes FIGS. 10A, B, and C shows raw BER (bit error rate) with an 8:12 modulation code before and after shift-compensation processing of the output signals from the detector pixels as a function of A) shift in the x-dimension, B) shift in the y-dimension, and C) shift along the x=−y diagonal.
A pixel shift-compensation algorithm is derived by considering the physical readout process used for holographic data storage. FIG. 3 shows a one dimensional (1-D) slice through three detector pixels 48 a, 48 b, 48 c that record and then read out respective optical intensities S0, S1, S2 to a processor 49. The processor may be a computer or a microcomputer, e.g., a microprocessor having dedicated hardware logic and/or software for performing the methods described herein. (Alternatively, the methods herein may be implemented off-line in a computer by reading out data from the processor 49 to the computer.) The three detector pixels 48 a, 48 b, 48 c are shown receiving optical beams (i.e., “bit images”) 50 a, 50 b, 50 c (having respective intensities p0, p1, p2) associated with SLM pixel elements 22 a, 22 b, 22 c. The SLM pixel elements 22 a, 22 b, 22 c are only included in FIG. 3 for illustrative purposes, as the SLM 18 need not be present during the readout process (see FIG. 2). Nevertheless, including the SLM pixel elements 22 a, 22 b, 22 c in FIG. 3 helps facilitate the discussion, since the reconstructed object beam 38 includes optical beams (such as 50 a, 50 b, 50 c) that are associated with the SLM pixel elements 22 as a result of the holographic storage process (see FIG. 1). The misregistration that occurs between the detector pixels 48 a, 48 b, 48 c and the optical beams 50 a, 50 b, 50 c may thus be thought of as a misregistration between the SLM pixel elements 22 a, 22 b, 22 c and the detector pixels 48 a, 48 b, 48 c.
In this example, the optical beams 50 a, 50 b, 50 c overlap with each other—more precisely, their respective electromagnetic fields 54 a, 54 b, 54 c interfere with each other at the detector pixels 48 a, 48 b, 48 c, and this interference necessitates deconvolving the output signals to determine the intensities of the individual beams 50 a, 50 b, 50 c. This interference is illustrated schematically in FIG. 4, in which the relative strengths of the electromagnetic fields 54 a, 54 b, 54 c at the detector pixels 48 a, 48 b, 48 c are shown as a function of position along the x-axis. The electromagnetic fields 54 a, 54 b, 54 c are seen in this example to be centered about the SLM pixel elements 22 a, 22 b, 22 c, but slightly offset from their respective detector pixels 48 a, 48 b, 48 c by an amount σx. This offset σx arises from the misregistration between the SLM pixel elements 22 a, 22 b, 22 c and the detector pixels 48 a, 48 b, 48 c in the x-direction; σx is the total misregistration in the x-direction arising from any source whatsoever, including optical distortion, magnification error, and material shrinkage. (See R. M. Shelby et al., “Distortions in pixel-matched holographic data storage due to lateral dimensional change of photopolymer storage media, Optics Letters, vol. 25, no. 10, pp. 713-715, May 15, 2000.) Thus, σx itself may vary slowly across the page.
In FIG. 4, the relative strengths of the electromagnetic fields 54 a, 54 b, 54 c at the detector pixels 48 a, 48 b, 48 c (as a function of position along the x-axis) are indicated in the z-axis for the sake of illustration. It will be understood by those skilled in the art that the amplitude of an electromagnetic field is a scalar quantity and has no particular orientation, even though the electromagnetic fields themselves may be thought of as propagating towards the detector pixels. Each of the electromagnetic fields 54 a, 54 b, 54 c has been “blurred” by the optical system, and this blurring is characterized by the optical system's point spread function (PSF). The relative electromagnetic field strength associated with each SLM pixel element 22, at the detector array 44, is denoted here by h(x) and is a function of position along the x-axis. The function h(x) can be calculated first by noting that the fill factor ffs for each of the SLM pixels elements 22 is constant. For convenience, a “4-F” (i.e., four focal lengths) system is considered as shown in FIGS. 1 and 2, in which the storage medium 10 is located at the back focal plane of a first lens 30 and at the front focal plane of a second lens 42, completing the imaging system from the SLM 18 to the detector array 44. Taking the Fourier transform aperture to be a square with sides of length D, h(x) is simply the convolution of the SLM pixel element and the sinc function PSF (see J. W. Goodman, Introduction to Fourier Optics, McGraw-Hill, 1968),
where sinc(x)=sin(Πx)/(Πx). Here, c is chosen such that h(x) is properly normalized:
Here DN=λf/δ is the Nyquist aperture defined in terms of the wavelength of the coherent radiation, the focal length of the optical system, and the pitch of the detector pixels 48. Note that position along the x-axis may be conveniently normalized to the pitch δ of the detector pixels 48, so that x is a measure of the number of pixels along the x-axis.
The signal intensities of the optical beams 50 a, 50 b, 50 c (associated with the SLM pixel elements 22 a, 22 b, 22 c, respectively), which are transmitted from the storage medium 10 to the detector array 44, are denoted by p0, p1, and p2, respectively. If the PSF is no broader than the pitch δ of the detector pixels 48, so that the amount of power lost into the second nearest pixel is insignificant, and if the offset σx between the SLM pixel elements 22 a, 22 b, 22 c and their corresponding detector pixels 48 a, 48 b, 48 c is non-zero, then it can be assumed that most of the signal detected by the detector pixel 48 c comes from the SLM pixel elements 22 b and 22 c. In this case, S2 is given by:
where ffd is the fill factor of the detector. As shown in Equation 3, optical intensity at any given point is given by summing all of the electromagnetic fields that contribute to the optical intensity at that point and then squaring the result. The total optical intensity S2 (which is measured by the pixel 48 c) can then be calculated through the appropriate integration.
Equation 3 gives a representation of the output S2 (which is uncorrected for misregistration) that is attributed to the pixels 22 b and 22 c for a known shift σx. The objective is to invert Equation 3, using the measured values S1 and S2 and the known shift σx, to compute p2. To solve Equation 3 for p2, it is convenient to introduce some abbreviations.
Equation 3 can be written as three separate terms:
Each of the three terms in Equation 4 represents a contribution to the measured signal S2 and is expressed as the product of an intensity (p1, p2) and a term Hij(σx) that arises from the fact that the optical system is misaligned. The first term represents the energy transmitted by the SLM pixel 22 c and received by the detector pixel 48 c, and is given by
The two digits in the subscript serve as a mnemonic that this is the electric field contribution from the SLM pixel having an associated offset of “0” (in this case, the pixel 22 c), multiplied by itself. In the absence of misregistration, this would be the only non-zero term. However, when there are misregistrations between the optical beams 50 a, 50 b, 50 c and the detector pixels 48 a, 48 b, 48 c, respectively, then a neighboring pixel will receive unintended signal: the electromagnetic field contribution from the SLM pixel having an associated offset of “1” (in this case, the pixel 22 b), multiplied by itself, is given by
This is the fraction of p1 from the optical beam 50 b inadvertently received through interpixel crosstalk at the pixel 48 c when the pixel 22 b is turned ON.
If the optical system used incoherent light, then the two terms given by Equations 5 and 6 would be the only ones required. One would find that one could solve explicitly for p2 by a linear equation involving S1 and S2, and it would be possible to incorporate a correction involving S3 (i.e., the signal originating from the pixel to the right of the pixel 48 c, which is not shown). However, with coherent light, the signal S2 is not simply the sum of the contribution due to the pixel 22 c and the contribution due to the pixel 22 b, but it also includes the indicated interference term. When p2 and p1 are 1.0, this interference term is
When the Fourier transform aperture is between 1 and 2 times the Nyquist aperture, then this integral is generally positive, implying that there is constructive interference between the beams 50 b and 50 c at the pixel 48 c. Since power must be conserved across the data page, expanding the limits of the integral in Equation 7 to infinity should result in zero net added signal. The constructive interference encountered at the pixel 48 c must thus be compensated by destructive interference at surrounding pixels. For instance, if H01 is positive, leading to an increase in S2, then S3, S1, and S0 may be correspondingly reduced. However, in the preferred embodiment, these second-order effects are ignored for simplicity and do not seem to have much impact on performance.
The interference between pixels can be other than constructive if the SLM pixels 22 a, 22 b, 22 c modulate not only the amplitude but also the phase of the electromagnetic field incident upon them. This may be done intentionally in order to distribute energy throughout the Fourier transform plane, with the idea that the phase will be unimportant to the intensity-based detectors. (This could be done with a programmable device capable of modulating both phase and amplitude, but using a fixed phase mask carefully aligned over the SLM 18 is more practical.)
In terms of Equation 4, one can consider the case in which the SLM 18 modulates the pixels with amplitude and phases a1exp(jφ1) and a2exp(jφ2), such that
The only modification to Equation 4 is then in its second term, so that S2 becomes
In the absence of a random phase mask, φ1 and φ2 are equal and the cosine term is 1, so that Equation 9 reduces to Equation 4. With a phase mask, the φ1-φ2 term may vary from pixel to pixel, so that some pixels experience constructive interference while others experience destructive interference. However, if the shift-compensation algorithm has knowledge of the original pattern of phase terms that were applied when the page was stored, then the appropriate terms can be applied in Equation 9.
An objective of the shift-compensation algorithm disclosed herein is to determine the intensities that would have been recorded by the detector pixels 48 a, 48 b, 48 c in the absence of misregistration. As described above, once the shift σx due to all contributing factors is known, the values of the three Hij terms can be computed. The value of S2 is the signal returned by the detector pixel 48 c and is a function of two unknowns: p2 and p1. If p1 is known, then Equation 4 can be used to solve for p2 and vice-versa, by writing it as a second-order polynomial in (p2)1/2:
and then solving for (p2)1/2 with the binomial equation, thereby yielding:
In order to use Equation 11, the value of p1 must already be known. An equation analogous to Equation 11 can be written for p1 in terms of S1 and p0, but at some point, the algorithm must be “seeded” with prior knowledge about one of the pixels in order to solve for the other intensities p1. One way to do this is to leave a blank pixel at the end of the row. Thus, if the pixel 22 a has been intentionally left OFF, then p0 is zero and p1 is simply S1/H00(σx). This value for p1 can then be substituted into Equation 11 to solve for p2. The shift compensation algorithm outlined here simply proceeds in an iterative fashion along an entire row, solving for the intensity that would have been recorded at each pixel in the absence of misregistration by using the calculated intensity of the just-processed, neighboring pixel. At each pixel, the algorithm starts with the measured signal Si, subtracts a portion that “belonged to” the previous pixel, subtracts a further portion due to interference, and then factors in the missing signal that should have been there but actually fell into the next pixel. While the algorithm requires a priori information about the shift σx and the associated weighting factors Hij(σx), no information about the binary pixel states or special encoding arrangement is required (other than that the pixel at the end of a row turned OFF, as discussed above). Note that the value of σx can change from pixel to pixel without loss of generality. (However, for each iterative step, only one value σx is used for the three terms in Equation 4.)
In FIG. 5, the detector pixels 48 a, 48 b, 48 c are shifted slightly leftward from the corresponding SLM pixels 22 a, 22 b, 22 c, and the SLM pixel element 22 a has been turned off, so that its corresponding electromagnetic field 54 a′ is essentially zero. The electromagnetic fields 54 b′ and 54 c′ are non-zero, however. The situation presented in FIG. 5 lends itself to solution using the methodology just outlined. In this case, the pixel 22 a has been set OFF, and p1, p2, and the calculated intensities corresponding to other detector pixels in the same row can now be determined. In the example of FIG. 5, the SLM pixel to the left of the pixel under consideration contributes most of the “crosstalk”. Accordingly, processing takes place from left to right.
FIG. 5 shows the left hand end of a row of pixels seeded on the left hand side with the pixel 22 a seeded OFF, and with processing taking place from left-to-right. In contrast, FIG. 6 shows the right hand end of a row of pixels, in which detector pixels 48 x, 48 y, 48 z are shifted slightly rightward with respect to the SLM pixels 22 x, 22 y, 22 z (having associated electromagnetic fields 54 x, 54 y, 54 z, respectively). In this case, σx is negative, and the SLM pixel to the right of the pixel under consideration contributes most of the significant crosstalk. Thus, it is preferable to process the row shown in FIG. 6 from right-to-left, using the following alternate equations for H11:
and for S2
Accordingly, the pixels at both ends of a row are preferably left OFF, one to seed the left-to-right pass (as discussed above) and one to seed the right-to-left pass.
If |σx|>0.5, then the term p1H11(σx) in Equation 4 becomes large, and the algorithm becomes prone to error-propagation. In this case, the correction factors to be applied to each pixel become large, and any random noise (arising from scattering or detector noise) from a given pixel is “amplified” along with the signal to affect the intensity calculated for the next pixel, leading to greater error as the algorithm proceeds iteratively along the row. For instance, when the pixel shift is small, perhaps 80% of the energy falls on the “correct” (intended) pixel and 20% on an adjacent, unintended pixel. Thus, if a signal of 100 (in units of the 8-bit gray-level camera signal) is received, the algorithm calls for increasing this by a factor of 1.25 (essentially “returning” the energy to the intended pixel from its neighboring pixel). If random noise results in a larger signal of 105, this factor of 1.25 increases the error in the calculation only slightly, to about 6 more than the corrected value of 125. However, if 20% of the energy falls on the intended pixel, and 80% on a neighboring pixel, the situation is quite different. Now the algorithm calls for a correction factor of 5, so that those 5 units of noise (when multiplied by 5) become 25 units. Error can thus propagate throughout subsequent pixel value calculations.
For this reason, it is preferred to process each row back and forth: once from left-to-right to compensate for rightward shifts of up to half a pixel as in FIG. 5, and once from right-to-left for leftward shifts as in FIG. 6. The signal recorded from each pixel in a row is processed on both “passes”, but only one of the two results is saved. Continuing the discussion related to the situation just described, imagine that the pixel 48 b loses only 20% of its energy to its neighboring pixel 48 c. On the left-to-right pass, the factor of 1.25 boosts the corrected signal from 100 to 125, and the pixel 48 c “loses” 25 counts of its signal (not considering any signal correction arising from the interference term between these 25 counts and the signal arising from the intensity p2.) However, on the right-to-left pass, the same pixel 48 b now encounters the other scenario, because it is the 20% of the signal that has been misplaced” into the pixel 48 c that is encountered first when iteratively calculating from right-to-left. After the signal from 48 c is processed on this right-to-left pass, there is only a small amount of leftover energy (the missing 20%), which is multiplied by a correction factor of 5 to estimate a corrected signal for the pixel 48 b. In this case, the value calculated for the intensity from the pixel 48 b in the right-to-left pass is discarded, while the value from the left-to-right pass is retained. Nevertheless, the pixel 48 b cannot be ignored on the right-to-left pass: The next pixel 48 a might have a smaller σx, and the resulting intensity for the pixel 48 b calculated by the iterative algorithm will seed that next pixel's intensity calculation.
Note that a pixel shift of exactly one pixel requires only bookkeeping (e.g., imagine that the optical beam 50 a from the pixel 22 a lands on the pixel 48 b, and the optical beam 50 b from the pixel 22 b lands on the pixel 48 c), and blank columns at the page boundaries can be used to prevent data from being lost off the edge of the data page. These blank columns also serve to seed the algorithm as discussed above. Further, for purposes of the shift compensation algorithm, a shift of 0.75 can be treated as a pixel shift of 1 pixel (which can be handled through bookkeeping) with σx set equal to −0.25. By extension, if the shift compensation algorithm works well for all shifts in the range −0.5<|σx|<0.5, then the algorithm is relatively insensitive to misalignment. Unfortunately, the transition at |σx|˜0.5 is problematic, especially if the estimate of the shift is flawed. The problem here is that the intensities calculated by the algorithm vary rapidly as a function of σx when |σx| is close to 0.5. For example, in the context of the 80%/20% scenario example above, a slightly different σx might lead to a division of energy of 75%/25%. The correction factor in one direction changes only from 1.25 to 1.333, while in the other direction it changes from 5 to 4. To reduce this effect, any pixel for which |σx|>0.4 is preferably processed on both the left-to-right and right-to-left passes, and a weighted average of the two estimates is made.
To extend the shift compensation algorithm to 2-D, one can extend the analysis of Equation 4 to three neighboring pixels (horizontal, vertical, and diagonal), producing an equation for calculated (i.e., corrected) intensity that has four linear and six nonlinear terms:
The four pixels in question, designated ps for “self”, ph for “horizontal”, pv for “vertical”, and pd for “diagonal”, are shown in FIG. 7, and the approach follows that outlined above: the signal received at a given pixel is a sum of linear contributions from the three neighbors, plus interference terms between all of the pairs of amplitudes, as suggested by the subscripts in Equation 15. Each Hij term is now a function of both the x and y offsets, σx and σy, respectively. FIG. 7A shows the case for which σx and σy are positive, so that pd sits above and to the left of ps, and left-to-right processing solves for the intensities of all three of pd, pv, and ph before ps is considered. FIGS. 7B, 7C, and 7D correspond to the other possible scenarios.
As with the 1-D variant of the algorithm, each of these Hij parameters can be calculated or measured by imaging known patterns of discrete optical beams from the SLM pixel elements 22 onto the detector pixels 48. For example, to measure Hss (σx, 0) for σx>0, one can measure the output from the first detector pixel to the right of the one for which the calculation is being made. In terms of FIG. 7A, this corresponds to the signal that would be received at Ss when only ph is turned on. (For σx<0, corresponding to FIGS. 7C and 7D, one can measure the output from the first pixel to the left of the one for which the calculation is being made.) To measure Hhs(σx,0) for σx>0, one can measure output from the detector pixel to the right of the SLM pixel illuminated above, but with the SLM pixel to the right also ON. This corresponds to the intensity received in Ss when both ps and ph are ON. If the system were linear, the signal received would now be identical to the sum of the Hss and Hhh measurements. Since the system is not linear, however, there is a difference that is quantified by Hsh. Care must be taken to make sure the factors of 2 in Equations 4 and 15 are neither ignored nor double-counted.
When the ten Hij parameters in Equation 15 were determined using the holographic storage platform known as “DEMON2” (see J. Ashley et al., supra; and G. W. Burr et al., “Volume holographic data storage at an areal density of 250 Gigapixels/in2”, Optics Letters, 26(7):444-446, 2001), it was found that all of these two-dimensional data sets could be separated into x- and y-dependent functions. (The methods herein are intended to work with any number of holographic platforms.) As is evident from Equations 1 and 5, these Hij functions arise from integrating the convolution of the SLM pixel shape and the PSF of the optical system. The PSF can either be dominated by diffraction or by aberrations. In the former case, the PSF will simply be the Fourier transform of the limiting spatial aperture at or near the storage medium 10. If this aperture is described by a separable function (e.g., a square or rectangle), then the PSF will be separable (the product of a sinc function in x and a sinc function in y). If aberrations dominate, however, the PSF will be further convolved with the effects of spherical aberration, coma, and other aberrations, and may change in shape and size for different positions in the data page.
The fact that the Hij functions were found to be separable and roughly equal regardless of where on the data page they were measured implies that diffraction, not aberrations, dominated the PSF in the DEMON2 platform used. For this reason, intensities from the detector pixels were corrected by employing only the 1-D algorithm. The 1-D algorithm was used first in the x-dimension, and only after these calculations were made the data page was reprocessed by applying the 1-D algorithm in the y direction. (Alternatively, the 1-D algorithm may be applied first in the y direction, and then in the x direction.) This procedure is possible when the 2-D functions involved can be separated into x and y components. Thus each 2-D data page was processed with the 1-D algorithm four times: twice horizontally (as described above) for all rows to produce an intermediate page (yielding intensities that were corrected for all x shifts σx), and then twice vertically (top-to-bottom and bottom-to-top) on all columns of this intermediate page. (The ends of the columns were seeded to zero, in analogy with the methodology presented for rows, as discussed above.) Using a computer program (compiled C code) on a 600 MHz Pentium III, this took 2.5 seconds for a 1024×1024 data page.
In FIG. 8A, a small 9×9 pixel block 60 a shows bright and dark pixels corresponding to pixel intensities as they should ideally be received. FIG. 8B shows a pixel block 60 b, which is the pattern of FIG. 8A imaged through the DEMON2 platform when the SLM is shifted a half-pixel in both x and y. The pixel block 60 b shows pixels having intermediate levels of brightness. The DEMON2 platform pixel-matches megapel pages through an aperture of 1.36DN (1.7×1.7 mm2 aperture, effective focal length of Fourier optics f=30 mm, wavelength λ=532 nm, 1024×1024 SLM with square pixels of pixel pitch δ=12.8 μm). In FIG. 8C, a pixel block 60 c is shown after the data of FIG. 8B has been processed with the shift-compensation algorithm to at least partially recover the information lost through misregistration. Note that the original binary pixel pattern could be accurately recovered by applying a single gray-level threshold to the grey pixels of FIG. 8C to distinguish bright and dark pixels. The very first implementation of the full 2-D algorithm described by Equation 15 using the DEMON2 holographic platform (note that the 2-D algorithm would have been mandated had aberrations dominated the PSF) gave results that were no better than those made with the 1-D algorithm. However, because the 1-D algorithm was simpler and worked quite well, it was sufficient to work with only the 1-D algorithm.
To process a block of pixels, the 1-D (or 2-D) algorithm needs only the total lateral shift, which is the sum of the dynamic misregistration and the static misregistration. The dynamic misregistration is that misregistration arising from errors in alignment affecting the entire detector array 44 (it is a “global” misregistration), e.g., the detector array may rest on a translation stage that is not brought into good alignment with the pixelated data page (propagating away from the SLM 18). Also, there may be changes between readouts due to mechanical vibrations or slow thermal drift of optical components. The static misregistration represents misregistration that is intrinsic to the optical system, e.g., offsets due to magnification error, material shrinkage, and optical distortion. In short, it is necessary to compensate not only for motion of the entire data page with respect to the detector array 44 (dynamic misregistration) but also for intrinsic distortions across the page (static misregistration). The dynamic shifts are determined between optical readouts by noting the location of dedicated fiducial marks within the pixelated data page and thus are updated every time a new page is read; the static misregistrations can be kept in a lookup table and are determined by calibration of the system. Such a calibration can be performed by detecting pages having known patterns of ON/OFF pixels under varying misregistrations, and for each small block of pixels within the data page (where a block represents, say, 10×10 pixels), selecting the σx, σy values that minimize the BER after processing.
FIG. 9 shows a data page 76 (which is the reconstructed object beam) divided conceptually into small blocks of data bits or “pixels” 78. Most of these blocks are data blocks 98 that are encoded to contain data and that are associated with an individual pair of static misregistrations (for the x and y dimensions, respectively) for that particular data block. In addition, the data page 76 includes fiducial blocks 70 strategically placed across the pixelated data page 76, such as the four blocks 70 shown in FIG. 9 arranged in the pattern of a square whose side is a reasonably large fraction of the page width. The data page 76, including the blocks 70, can be thought of as a page of pixels 78 corresponding to the SLM pixel elements 22 (which are associated with discrete optical beams such as 50 a, 50 b, 50 c). A portion 80 of the data page 76 is dedicated to pixels having encoded data, whereas another portion is set aside for fiducial purposes. The fiducial portion (comprising the blocks 70) includes isolated on-pixels 84 surrounded by pixels 86 that are OFF. Thus, the fiducial blocks 70 are made up mostly of OFF (dark) pixels 86 and a few, isolated ON pixels 84 widely separated from each other. This wide separation allows the optical beams associated with each of the pixels 84 propagating onto the detector array 44 to be computed by determining the center of the light falling onto the detector pixels 48 at and around the “target” detector pixel (i.e., the pixel that would, in the absence of misregistration, receive all the light from its respective ON pixel). This determines the translation of the data page 76 relative to the detector array 44. If detector pixels to the left and to the right of the target pixel receive the same optical intensity from an ON pixel 84, then that ON pixel is taken to be centered in the x-direction; a weighted average can be used for this purpose. An analogous procedure can be used for the y-direction.
The OFF pixels 86 guarantee that when the centroid of an discrete optical beam is computed, only light originating from a single pixel in the SLM is used, so that neighboring pixels do not contribute signal through interpixel crosstalk. Each fiducial block 70 thus provides an x and a y offset in units of pixels, and these offsets can be averaged over the fiducial blocks on the page to determine the dynamic misregistration of the data page 76 with respect to the detector array 44.
On the other hand, the static misregistrations between each data block 98 and its associated detector pixels 48 need to be calibrated for each block 98 of data pixels. These static misregistrations can be thought of as the x, y offsets associated with the best alignment position for that particular block. Thus, if a particular block 98 has a static misregistration or “local” x, y offset of (0.1, 0.2), this implies that this data block would have been perfectly aligned with its intended detector pixels 48 if the dynamic (global) misregistration had been (−0.1, −0.2). For a dynamic misregistration of (0.4, 0.3), as determined by the fiducial blocks 70, this same block would be misaligned by exactly half a pixel in both the x and y dimensions. A set of total misregistrations σx, σy, between the detector elements 48 and the pixels 78 of the data page 76 (corresponding to the SLM pixel elements 22 and their associated optical beams, such as 50 a, 50 b, 50 c) may be computed by adding the dynamic misalignment to the static misregistrations.
In one implementation of the algorithm, pairs of static (local) x and y offsets, σx and σy, respectively, with a resolution of 0.025 pixel, were calculated for each block of 10×10 pixels across a megapel data page. Thus, the Hij lookup tables contained one entry for each 0.025 pixel increment, requiring 21 values to cover σx,y from 0 to 0.5. No values are expressly listed to cover 0 to −0.5 because the left-to-right or right-to-left nature of the pass already accounts for which pixel the shift correction is applied to. In other words, the H01 (σ=0.25) value used in Equation 4 is identical to the H01(σ=−0.25) value used in Equation 14. One set of Hij (σx,y) values was used across the entire page, measured as described above. Values for the row-by-row correction of x shifts based on σx, and values for the column-by-column correction of y shifts based on σy were stored separately. Note that it might be expected that with a square Fourier transform aperture and square SLM pixels, the system should be symmetric. However, the Hij(σx) and Hij(σy) values were measured and found to be different in this particular implementation. This implies that the detector pixels 48 were not exactly square for the CCD detector array 44 in this particular implementation. (Evidently, the symmetry of the detector pixels 48 was intentionally broken by the designer in order to accommodate improved CCD readout.) This makes the value of ffd used in Equations 5-7 different for the x and y dimensions. Thus, for each dimension, there were three lookup tables, corresponding to H00, H01, and H11, each with 21 unique values to cover 0<|σ|<0.5.
FIGS. 10A, B, and C show the resulting page-wide performance for images transmitted through the DEMON2 system as a function of x misalignment, y misalignment, and for the case when the misalignment in the x and y dimensions are equal. Similar results have been demonstrated with holograms stored at areal densities up to 250 Gbit/sq. in. (See G. W. Burr et al., “Volume holographic data storage at an areal density of 250 Gigapixels/in2”, Optics Letters, 26(7):444-446, 2001). The raw-BER (bit error rate) was measured, both before and after applying the shift-compensation algorithm, at the output of an 8-bits-from-12-pixels modulation decoder into which signal outputs were fed. (For a discussion of modulation decoding, see G. W. Burr, J. Ashley, H. Coufal, R. K. Grygier, J. A. Hoffnagle, C. M. Jefferson, and B. Marcus, “Modulation coding for pixel-matched holographic data storage”, Optics Letters, 22(9):639-641, 1997.) As expected, the BER increases when either |σx| or |σy| is near 0.5, and decreases again for pixel shifts close to ±1. Assuming that 10−3 is the maximum acceptable raw-BER that can be corrected by error-correction codes (see M. A. Niefeld et al., “Error correction for increasing the usable capacity of photorefractive memories”, Optics Letters, vol. 19, no. 18, pp. 1483-1485, 1994), FIG. 10 shows that the position tolerance in x and y for this particular implementation increases from ±16% to ±40% of the pixel pitch when the shift-compensation algorithm is applied. For example, without the algorithm, the x or y alignment must be within ±16% of the pixel pitch, or 1.92 microns when the pixel pitch is 12 microns. By using the compensation algorithm procedure described herein, greater misalignment is possible while still getting acceptable BER.
As discussed above, one of the factors leading to high BER at |σ|˜0.5 is error-propagation, since noise-induced errors in the signal have a strong influence on the calculated intensities. One way to reduce this error is to insert rows and columns of OFF pixels within the data page 76 so that the algorithm can re-seed itself, thereby ensuring that errors from one pixel do not propagate across the page as the algorithm is iteratively carried out from pixel to pixel. Since these OFF pixels cannot encode data, this trades off code rate for improved performance. Inserting rows and columns of OFF pixels also permits parallel execution, reducing the buffer size and processing latency. Alternatively, one could intentionally introduce a small (˜1%) magnification error, so that σx and σy vary quickly through |σ|˜0.5 in small isolated patches. This decreases the length of continuous pixel sections with large correction factors, thus reducing error propagation.
Additionally, Equation 4 could simply be used as an improved channel model within existing parallel detection (see B. M. King et al., 1998, supra) and sequence-estimation schemes. This reference describes post-processing algorithms that try to determine the data patterns originally stored with an SLM by comparing the measured data against a modeled prediction of what each possible stored data pattern would have produced. In this context, Equation 4 represents an improved model for generating candidate distorted outputs for comparison with the actual measured output. Such algorithms typically choose the input that leads to the distorted output most closely resembling the measured data, so that a model that more closely follows the actual physics can offer over improved performance over the inaccurate but simple linear channel model.
In one embodiment of the invention, there is provided media encoded with executable program code to effect any of the methods described herein. These media may include a magnetic or optical disk or diskette, for example. In preferred embodiments, this program code may be read by a digital processing apparatus such as a computer for performing any one or more of the methods disclosed herein.
The invention may be embodied in other specific forms without departing from its spirit or essential characteristics. The described embodiments are to be considered in all respects only as illustrative and not restrictive. The scope of the invention is therefore indicated by the appended claims rather than the foregoing description. All changes within the meaning and range of equivalency of the claims are to be embraced within that scope.
|Cited Patent||Filing date||Publication date||Applicant||Title|
|US3715733 *||Feb 11, 1972||Feb 6, 1973||Bell Telephone Labor Inc||Error compensating optical detector apparatus and method|
|US5379266 *||Dec 30, 1991||Jan 3, 1995||Information Optics Corporation||Optical random access memory|
|US5511058||Dec 23, 1993||Apr 23, 1996||Tamarack Storage Devices||Distortion correction of a reconstructed holographic data image|
|US5694488||Feb 21, 1996||Dec 2, 1997||Tamarack Storage Devices||Method and apparatus for processing of reconstructed holographic images of digital data patterns|
|US5838650||Jun 26, 1996||Nov 17, 1998||Lucent Technologies Inc.||Image quality compensation method and apparatus for holographic data storage system|
|US5940537||Jan 16, 1996||Aug 17, 1999||Tamarack Storage Devices||Method and system for compensating for geometric distortion of images|
|US5978112 *||Nov 25, 1996||Nov 2, 1999||California Institute Of Technology||Non-volatile readout of shift multiplexed holograms|
|US5982513 *||Feb 9, 1998||Nov 9, 1999||Holoplex, Inc.||Method and system to align holographic images|
|US6111776 *||Aug 27, 1998||Aug 29, 2000||Ioptics, Incorporated||Compact optical random access memory having multiple reflections|
|US6414296 *||Sep 21, 2000||Jul 2, 2002||Imation Corp.||Multiple pixel reading of holographic data including a position error calculation|
|1||"Post-Processing to Correct for Optical Distortion and Material Shrinkage in Holographic Data Storage", SPIE's International Technical Group Newsletter (published on the WEB), Holography, Dec. 2000, p. 4.|
|2||B. M. King, et al., "Parallel Detection Algorithm for Page-Oriented Optical Memories," Applied Optics, vol. 37, No. 26, Sep. 10, 1998, pp. 6275-6298.|
|3||F. Zhao, et al., "High Density Phase-Conjugate Holographic Memory with Phase-Only Image Compressors," Optical Memory and Neural Networks, vol. 6, No. 4, 1997, pp. 261-264.|
|4||G. W. Burr, et al., "Compensation for Pixel Misregistration in Volume Holographic Data Storage," Optics Letters, vol. 26, No. 8, Apr. 15, 2001, pp. 542-544.|
|5||G. W. Burr, et al., "Compensation of Pixel Misregistration for High Density Volume Holographic Data Storage," Abstracts of OSA Annual Conference available on WEB, Aug. 2000.|
|6||G. W. Burr, et al., "Modulation Coding for Pixel-Matched Holographic Data Storage," Optics Letters, vol. 22, No. 9, May 1, 1997, pp. 639-641.|
|7||G. W. Burr, et al., "Volume Holographic Data Storage at an Areal Density of 250 Gigapixels/in.2,"Optics Letters, vol. 26, No. 7, Apr. 1, 2001, pp. 444-446.|
|8||J. Ashley, et al., "Holographic Data Storage," IBM J. Res. Development, vol. 44, No. 3, May 2000, pp. 341-368.|
|9||J. F. Heanue, et al., "Signal Detection for Page-Access Optical Memories with Intersymbol Interference," Applied Optics, vol. 35, No. 14, May 10, 1996, pp. 2431-2438.|
|10||M. A. Neifeld, et al., "Error Correction for Increasing the Usable Capacity of Photorefractive Memories", Optics Letters, Sep. 15, 1994, vol. 19, No. 18, pp. 1482-1485.|
|11||M. Bernal, et al., "Balancing Interpixel Cross Talk and Detector Noise to Optimize Areal Density in Holographic Storage System," Applied Optics, vol. 37, No. 23, Aug. 10, 1998, pp. 5377-5385.|
|12||R. M. Shelby, et al., "Distortions in Pixel-Matched Holographic Data Storage Due to Lateral Dimensional Change of Photopolymer Storage Media," Optics Letters, May 15, 2000, vol. 25, No. 10, pp. 713-715.|
|13||R. M. Shelby, et al., "Pixel-Matched Holographic Data Storage with Megabit Pages," Optics Letters, vol. 22, No. 19, Oct. 1, 1997, pp. 1509-1511.|
|14||V. Vadde et al, "A Figure-of-Merit for the Optical Aperture Used in Digital Volume Holographic Data Storage," Optical Data Storage '98, Proceedings of SPIE, vol. 3401, pp. 194-200.|
|15||V. Vadde et al, "Channel Estimation and Intra-Page Equalization for Digital Volume Holographic Data Storage," Optical Data Storage '97, Proceedings of SPIE, Vol. 3109, pp. 245-250.|
|16||V. Vadde, et al., "Channel Modeling and Estimation for Intrapage Equalization in Pixel-matched Volume Holographic Data Storage," Applied Optics, vol. 38, No. 20, Jul. 10 1999, pp. 4374-4386.|
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7206105 *||May 19, 2003||Apr 17, 2007||Pioneer Corporation||Hologram recording device and hologram reproduction device|
|US7321541 *||Oct 9, 2003||Jan 22, 2008||Optware Corporation||Information recording method, reproducing method, and recording/reproducing method utilizing holography|
|US7369962 *||Jan 6, 2005||May 6, 2008||Micronic Laser Systems Ab||Method and device for data integrity checking|
|US7519224 *||Apr 6, 2005||Apr 14, 2009||Daewoo Electronics Corporation||Method and apparatus for compensating data image reproduced from holographic system|
|US7570838||Dec 30, 2004||Aug 4, 2009||Daewoo Electronics Corporation||Image correction method for compensating for misalignment of pixels in holographic digital data storage system|
|US7653248 *||Jan 26, 2010||Science Applications International Corporation||Compression for holographic data and imagery|
|US7738736 *||May 6, 2004||Jun 15, 2010||Inphase Technologies, Inc.||Methods and systems for holographic data recovery|
|US7757095 *||Jan 6, 2005||Jul 13, 2010||Tdk Corporation||Personal identification method, personal identification system, and optical information recording medium|
|US7782732 *||Aug 24, 2010||Fuji Xerox Co., Ltd.||Optical recording method, optical reproducing method, and optical reproducing apparatus|
|US7787346||Jun 12, 2007||Aug 31, 2010||Fuji Xerox Co., Ltd.||Hologram recording method and device, hologram reproduction method and device, and optical recording medium|
|US7848595||Dec 7, 2010||Inphase Technologies, Inc.||Processing data pixels in a holographic data storage system|
|US7916958||Mar 29, 2011||Science Applications International Corporation||Compression for holographic data and imagery|
|US8000204 *||Aug 16, 2011||Lite-On It Corporation||Data processing method for a holographic data storage system|
|US8135219||Mar 7, 2007||Mar 13, 2012||Lite-On It Corp.||Method of detecting and compensating fail pixel in hologram optical storage system|
|US8275216 *||Jun 28, 2004||Sep 25, 2012||Inphase Technologies, Inc.||Method and system for equalizing holographic data pages|
|US8290254 *||Oct 16, 2012||National Chiao Tung University||Gray level weighting centroid method for holographic data storage|
|US9305592 *||Dec 24, 2014||Apr 5, 2016||Foundation Of Soongsil University-Industry Cooperation||Reception terminal and a method for compensating inter-symbol interference and computer readable recording medium for performing the same|
|US20050018263 *||May 6, 2004||Jan 27, 2005||Pharris Kenton J.||Methods and systems for holographic data recovery|
|US20050177326 *||Jan 6, 2005||Aug 11, 2005||Micronic Laser Systems Ab||Method and device for data integrity checking|
|US20050185230 *||May 19, 2003||Aug 25, 2005||Pioneer Corporation||Hologram recording device and hologram reproduction device|
|US20050226528 *||Apr 6, 2005||Oct 13, 2005||Daewoo Electronics Corporation||Method and apparatus for compensating data image reproduced from holographic system|
|US20050286095 *||Dec 30, 2004||Dec 29, 2005||Daewoo Electronics Corporation||Image correction method for compensating for misalignment of pixels in holographic digital data storage system|
|US20050286387 *||Jun 28, 2004||Dec 29, 2005||Inphase Technologies, Inc.||Method and system for equalizing holographic data pages|
|US20050286388 *||Feb 28, 2005||Dec 29, 2005||Inphase Technologies, Inc.||Processing data pixels in a holographic data storage system|
|US20060050544 *||Oct 9, 2003||Mar 9, 2006||Optware Corporation||Information recording method, reproducing method, and recording reproducing method utilizing holography|
|US20070236763 *||Nov 27, 2006||Oct 11, 2007||Fuji Xerox Co., Ltd.||Optical recording method, optical reproducing method, and optical reproducing apparatus|
|US20070288757 *||Jan 6, 2005||Dec 13, 2007||Takuya Tsukagoshi||Personal Identification Method, Personal Identification System, And Optical Information Recording Medium|
|US20080033677 *||Jun 11, 2007||Feb 7, 2008||Applera Corporation||Methods And System For Compensating For Spatial Cross-Talk|
|US20080062488 *||Mar 7, 2007||Mar 13, 2008||Lite-On It Corp.||Method of detecting and compensating fail pixel in hologram optical storage system|
|US20080151734 *||Jun 12, 2007||Jun 26, 2008||Fuji Xerox Co., Ltd.||Hologram recording method and device, hologram reproduction method and device, and optical recording medium|
|US20080244371 *||May 5, 2008||Oct 2, 2008||Micronic Laser Systems Ab||Method and device for data integrity checking|
|US20080316895 *||Jun 6, 2008||Dec 25, 2008||Lite-On It Corporation||Data Processing Method for a Holographic Data Storage System|
|US20090141120 *||Nov 17, 2008||Jun 4, 2009||Daewoo Electronics Corporation||Method and apparatus for compensating data image reproduced from holographic system|
|US20100046848 *||Feb 25, 2010||Hanna Elizabeth Witzgall||Compression For Holographic Data and Imagery|
|US20100061213 *||Sep 26, 2007||Mar 11, 2010||Koninklijke Philips Electronics N.V.||Optical holographic device and method with gain compensation|
|US20110182505 *||Jul 28, 2011||National Chiao Tung University||Gray level weighting centroid method for holographic data storage|
|CN1305043C *||Jun 18, 2004||Mar 14, 2007||清华大学||Pixel matching real-time compensation method for position multiplexing volume holographic data storage system|
|CN101783156B||Dec 16, 2009||Jun 18, 2014||通用电气公司||Method and system for optical data storage with enhanced detection|
|WO2008002765A2 *||Jun 11, 2007||Jan 3, 2008||Applera Corporation||Method and system for compensating for spatial cross-talk|
|U.S. Classification||369/103, 365/125, 365/216, G9B/7.027|
|International Classification||G11B7/00, G11B7/0065|
|May 1, 2001||AS||Assignment|
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, NEW Y
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:BURR, GEOFFREY W.;REEL/FRAME:011785/0001
Effective date: 20010501
|Sep 14, 2004||CC||Certificate of correction|
|Jul 13, 2007||FPAY||Fee payment|
Year of fee payment: 4
|Jul 15, 2011||FPAY||Fee payment|
Year of fee payment: 8
|Oct 2, 2015||REMI||Maintenance fee reminder mailed|
|Feb 24, 2016||LAPS||Lapse for failure to pay maintenance fees|
|Apr 12, 2016||FP||Expired due to failure to pay maintenance fee|
Effective date: 20160224